AI that actually works. Not somewhere in theory, but in the day-to-day business of a Swiss company. In this update, we show how ASTORtec reduced manual order entry to a minimum with a custom AI solution from Evoya AI.
Also: what's new on the Evoya platform since June – from new models to entirely new features.
- Customer project ASTORtec: PDF in, ERP entry out – how order entry was automated.
- Platform update: Files section, Projects, Agent Builder, Feature Access, new AI and image models and more.
PDF in, ERP entry out: how ASTORtec automated order entry
Check every order by hand, type it up, transfer it to the ERP. Per order: around 10 minutes. Considerably more for extensive PDFs. That was everyday life at ASTORtec, a Swiss technical trading company that receives B2B orders by email as PDF attachments every day.
The problem: every customer sends a different format. Individual article numbers, different units of measure, later corrections. Everything has to be recognised, matched and transferred correctly into the system. Error-prone, time-consuming and hard to scale.
The solution: the Evoya Order App. Together with ASTORtec, Evoya AI developed a custom application that automates this process – from the incoming email to the finished ERP entry in Europa3000.
How it works, in four steps:
- Detect incoming orders & read the PDFs: Mistral OCR extracts text, tables and line items from the attachments.
- Interpret & structure the data: language models bring the order contents into a uniform structure; customer-specific prompts account for individual formats and article logic.
- Validate & hand over directly to the ERP: defined rules check quantities, delivery dates and article mapping. Green light: the order is created automatically in Europa3000.
- Hold back exceptions for manual approval: anything ambiguous lands in the web interface for review – already prepared, no longer from scratch.
The result? Orders that used to take 10 minutes of manual work are now processed automatically or reviewed and approved in 1 to 2 minutes.
On data protection: the order entry application developed by Evoya AI is hosted on Swiss infrastructure. For the AI processing – reading and interpreting the PDFs – the solution relies on Mistral AI, operated on sovereign EU infrastructure. For projects with higher requirements on where AI processing takes place, Evoya AI also offers Swiss solutions or on-premises operation.
And the principle isn't limited to orders: delivery notes, invoices, quotes, forms. Wherever information from PDFs has to flow into business systems, the same approach applies.
Read all the details in the full case study. Do you have similar processes in your company? Talk to us – together we'll look at what can be automated.
What's happened on the Evoya platform since June
A lot. Really a lot. Here are the highlights at a glance, with direct links to the details in the changelog.
- New Files section (23 September): find and manage files and images faster in one place. A clear folder structure, direct uploads and an image library that shows all generated images at a glance. Learn more
- GLM 5.3 on sovereign EU infrastructure (22 September): the open-source model GLM 5.3 from Z.ai is now available in its Mistral-hosted version. Strong with long texts (1 million token context), coding and agentic workflows, at considerably lower cost than comparable models. Learn more
- New image models & improved image prompt (21 September): two new image generation models: GPT Image 2.5 Flare for fast, high-quality everyday generation and GPT Image 2.5 Sunburst for maximum quality and precise image editing. New: simply state the format you want (landscape, portrait, square) directly in the prompt. Learn more
- New OCR service from Mistral AI (18 September): faster and more accurate text recognition from images and documents, including tables, forms and handwriting. Runs on sovereign European infrastructure, fully compliant with the Swiss FADP and the GDPR. Learn more
- New AI models (17 September): five new models on the platform: GPT-6 Astra and Claude Fable 5.1 for maximum quality, Gemini 3.8 Flash as a strong default, and GLM 5.3 Flash and DeepSeek v4.1 Flash for cost-efficient work at high volume. Learn more
- New OpenAI models (25 August): GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna are now available on the platform – ready to use in agents and workflows. Learn more
- Feature Access (21 August): not every user needs every feature. Admins now control per role which features are visible and accessible. The platform stays clear and tailored to each team's needs. Learn more
- Projects in scheduled tasks (20 August): scheduled tasks can now be linked to a project folder. Daily reports, created automatically and filed in the right place. No more scattered chaos. Learn more
- Folder-based data sources (14 August): data sources can now be linked to a folder. Drop new files into the folder and the data source updates automatically. No more manual uploads. Learn more
- Projects (27 July): familiar from ChatGPT or Claude, now on the Evoya platform: Projects give an agent its own workspace with file access, context and notes. Ideal for more complex, multi-step tasks. Learn more
- Agent Builder (18 June): create an AI agent in five guided steps, no technical knowledge needed. Just click "Create new agent" in the agent overview and get started. Learn more
You'll find all changes in detail in the full changelog.
Ready for more efficiency?
The Evoya platform is just a click away. Start a free trial or book a demo – we'll be happy to show you the features!